Minutes_2020_05_26

Numba Meeting: 2020-05-26

Attendees: Siu, Aaron, Graham, Stuart, Todd, Val, Hameer, Pearu

0. Feature Discussion

• CoC Draft discussion:
  • current PR https://github.com/numba/numba-governance/pull/1
  • let's get this in by end of week.
• RC1 release
  • pending LLVM9 buildfarm
  • error message improvement, inliner, entry-points features
  • reintroduce refct pruner pass with very conversative checking
• Review current cycle: 4 weeks devs + 2 weeks burndown
  • keep doing for another cycle
  • this cycle is unusual due fallout from prev release
  • burndown is good
• Performance measure
  • context: avoid major regression related to removing refcount pruner
  • performance testing using buildfarm and post to the PullRequest
  • also to think about:
    • revive existing benchmark as the quickest way to start measuring again https://github.com/numba/numba-benchmark
• Hameer interested to work on https://github.com/numba/numba/issues/5750

1. New Issues

• #5746 - Significant performance regression for datashader in numba 0.49.x
  • try to reintroduce the pruner but make it very strict to the situation that it would apply
• #5744 - Issue with vectorize for GPU
• #5743 - Encountered weird drop in performance when running a function
• #5742 - Runtime measurements
• #5739 - Bug using += (Add AND) with matrix transpose
  • likely overwritten same memory
• #5736 - np.isnan for np.datetime64 (numpy 1.18)
• #5735 - failure to use Prange with @njit(parallel=True) on loop which can be parallelised
  • need fix in array_analysis
• #5731 - Bug? Variable np.searchsorted side argument
• #5730 - Using jitclass with types.List of array results in TypingError
  • we should fix by implementing types.int64[:]
• #5728 - Var error in parfors step
  • there's a WIP fix
• #5727 - complex power fails with cuda
• #5726 - jitclass doesn't respect ndarray.flags.writeable flag
• #5723 - Overload float(), int() and str() to do number <-> string conversion
• #5720 - vectorize(target="parallel") breaks type resolution

Closed Issues

• #5737 - Numba CUDA kernel very slow compared to CuPy fuse

2. New PRs

• #5745 - CUDA stream callbacks and async awaitable streams
• #5741 - CUDA: Add two-argument implementation of round()
• #5740 - Add FAQ entry on how to create a MWR.
• ** #5734 - Add printf and fflush
  • wonder about usecase: it's for debugging
• #5732 - CUDA Docs: document forall method of kernels
• #5729 - Enable dispatcher instances as 1st class function types when used in argument tuples
• #5724 - Initialize process lock lazily to prevent multiprocessing issue
• #5722 - Improve error messages.
• #5721 - Fix array iterator layout.

Closed PRs

• #5738 - Unbreak master on buildfarm
• #5733 - Add np.isreal and np.iscomplex support.
• #5725 - don't merge! Wip/declarative typing 4
• #5719 - [WIP] Fix #5713: cache typed dict and list functions

3. Next Release: Version 0.50.0, RC=May 28th

• Requests for 0.50

• high risk stuff for 0.50.

  • Declarative typing (Siu and Stu)
    • declaring accepted types in things like @overload
    • so we can have better errmsg
    • allow overloading "contextual information"; i.e. targets (CPU, GPU), fastmath flag, exception handling
    • make sure we don't break existing @overload/@lower_builtin use cases
  • Remove macro expansion (Graham)
    • replace "macros" with standard typing/lowering; i.e. @register / @lower (@overload not yet working with CUDA)

• llvm 9

• plan to have first open meeting on June 4th.

• 0.51 potential tasks

  • TypedSet
    • base on TypedDict

    • Related:

    https://github.com/numba/numba/pull/5639
  • Nested modes
  • Move SSA pass up the pipeline
    • make sure more passes can work in SSA form
  • CUDA Dispatcher / kernel interface:
    • Share machinery with CPU target (e.g. for multiple signatures, typeconv, etc.)

4. Upcoming tasks

• Opening up the numba meeting
  • needs to approve a CoC
  • plan to have first open meeting on June 2nd.